Will AI take over the jobs of doctors who interpret X-rays, or will it simply enhance their abilities?
In Washington, there’s a burning question on the minds of many workers as ChatGPT and similar AI tools that can chat, spin yarns, and even whip up tunes and artwork within seconds. For medics poring over scans to detect cancer and other ailments, the shadow of AI has been present for around ten years, with an increasing number of algorithms vowing to boost precision, speed up tasks, and sometimes assume full control of certain duties.
The forecasts have been wildly varied, from grim prophecies where AI completely supplants radiologists to optimistic visions where it liberates them to concentrate on the most fulfilling parts of their profession. This dichotomy mirrors the broader rollout of AI in healthcare.
Beyond the technology itself, much depends upon the willingness of doctors to put their trust — and their patients’ health — in the hands of increasingly sophisticated algorithms that few understand. Even among experts, there’s debate over how eagerly radiologists should be adopting this tech.
“Some of the AI techniques are so good, frankly, I think we should be doing them now,” declared Dr. Ronald Summers, a radiologist and AI researcher at the National Institute of Health in the US. “Why are we letting that information just sit on the table? “.
Dr Summers’ lab has pioneered the development of computer-aided imaging programs that can detect a range of conditions including colon cancer, osteoporosis, and diabetes. However, these have not been widely adopted, something he attributes to the “culture of medicine,” among other factors.
Radiologists have been utilising computers to enhance images and highlight suspicious areas since the 1990s. But the latest AI programs are capable of much more, interpreting scans, offering diagnoses and even drafting written reports about their findings.
These algorithms are often trained on millions of X-rays and other images gathered from hospitals and clinics. The Food and Drug Administration (FDA) has approved over 700 AI algorithms to assist physicians across the medical field. More than 75% of these are in radiology, yet only an estimated 2% of radiology practices utilise such technology.
Despite the industry’s promises, radiologists have several reasons to be wary of AI programs: limited testing in real-world settings, lack of transparency about how they function and questions about the demographics of the patients used to train them.
Dr. Curtis Langlotz, a radiologist who runs an AI research centre at Stanford University, said: “If we don’t know on what cases the AI was tested, or whether those cases are similar to the kinds of patients we see in our practice, there’s just a question in everyone’s mind as to whether these are going to work for us,”.
So far, all the programs cleared by the FDA require human involvement. In early 2020, the FDA convened a two-day workshop to deliberate on algorithms capable of functioning autonomously. However, radiology experts swiftly cautioned the regulators in a letter, stating they “strongly believe it is premature for the FDA to consider approval or clearance” for such autonomous systems.
Despite this, European authorities gave the green light in 2022 to the first fully automated software that can assess and generate reports for chest X-rays deemed healthy and normal. The app’s creator, Oxipit, is now gearing up to seek FDA approval in the US. The demand for this tech is high in Europe due to some hospitals grappling with massive scan backlogs, exacerbated by a shortage of radiologists.
In the US the adoption of such automated screening could take years. This delay isn’t due to a lack of technological readiness, AI industry leaders suggest, but rather because radiologists are hesitant to hand over even basic tasks to machines.
“We try to tell them they’re overtreating people and they’re wasting a ton of time and resources,” said Chad McClennan, chief executive of Koios Medical, which sells an AI tool for ultrasounds of the thyroid, the vast majority of which are not cancerous. “We tell them, ‘Let the machine look at it, you sign the report and be done with it.’”
Radiologists often overestimate their own accuracy, according to Mr McClennan. His company’s research found that doctors examining the same breast scans disagreed with each other more than 30% of the time on whether a biopsy was necessary. The same radiologists contradicted their initial assessments 20% of the time when reviewing the same images a month later.
The National Cancer Institute reports that about 20% of breast cancers are overlooked during routine mammograms. Moreover, there’s a significant potential for cost savings. On average, US radiologists earn over $350,000 annually, according to the Department of Labor.
In the short term, experts predict AI will function like autopilot systems in aircraft – carrying out crucial navigation tasks but always under human supervision. This approach provides reassurances to both radiologists and patients, says Dr. Laurie Margolies from the Mount Sinai hospital system in New York.
The system utilises Koios breast imaging AI for a second opinion on mammography ultrasounds. “I will tell patients, ‘I looked at it, and the computer looked at it, and we both agree,'” Margolies stated. “Hearing me say that we both agree, I think that gives the patient an even greater level of confidence.”
The first large-scale, rigorous trials comparing AI-assisted radiologists with those working independently offer glimpses of potential improvements. A Swedish study of 80,000 women has revealed that a single radiologist working with artificial intelligence (AI) detected 20% more cancers in mammograms than two radiologists working without the technology.
In Europe, it’s standard practice for two radiologists to review mammograms to ensure accuracy. However, Sweden, like many other countries, is grappling with a workforce shortage, with only about 70 breast radiologists serving a population of 10 million. The study found that using AI instead of a second reviewer reduced the human workload by 44%.
Despite these promising results, the study’s lead author, Dr. Kristina Lang of Lund University, insists that a radiologist must always make the final diagnosis. If an automated algorithm misses a cancer, “that’s going to be very negative for trust in the caregiver.”
The question of who would be held liable in such cases is among the thorny legal issues that have yet to be resolved. As a result, radiologists are likely to continue double-checking all AI determinations to avoid being held responsible for any errors. This could negate many of the predicted benefits of AI, including reduced workload and burnout.
Dr. Saurabh Jha of the University of Pennsylvania believes that only an extremely accurate and reliable algorithm would allow radiologists to truly step away from the process. Until such systems emerge, Dr Jha likens AI-assisted radiology to someone who offers to help you drive by looking over your shoulder and constantly pointing out everything on the road.
“That’s not helpful,” says Dr Jha, “If you want to help me drive then you take over the driving so that I can sit back and relax.”
